Impact of follow-up on generalized pairwise comparisons for estimating the irinotecan benefit in advanced/metastatic gastric cancer

Background and objectives: The net treatment effect (∆) is a new method to assess the treatment benefit that combines multiple time-to-event, binary and continuous endpoints according to a pre-specified sequence. It represents the net probability for a random patient treated in the experimental arm to have a better overall outcome than a random patient from the control arm does. We aimed at characterizing the impact of follow-up on ∆ estimated from both time-to-event and binary toxicity endpoints, in randomized controlled trials (RCTs) of irinotecan-based regimen in advanced/metastatic gastric cancer (AGC). Study design: Three RCTs are reanalysed. The net treatment effect using from one to three outcomes (i.e. overall survival, time to progression and toxicity in this order) and the hazard ratio (HR) were estimated after various cut-off dates and compared to the values obtained after complete follow-up were reported. Results: In all three RCTs (897 patients), the irinotecan-based regimen was superior to the non-irinotecan containing regimen in terms of HR and ∆. This superiority was lower when the net treatment effect also accounted for toxicity. The HR was slightly less influenced by an incomplete follow-up than ∆ was, but correction proposed by Péron to account for censored observations showed quite robust results. Conclusions: The net treatment effect using Péron's correction can be used in case of interim analyses or high censoring rates. In addition to relative measures such as the hazard ratio, it provides a simple mean to evaluate the net treatment effect with and without toxicity outcomes.